Research report
The relationship between active ageing and health using longitudinal data from Denmark, France, Italy and England Giorgio Di Gessa,1 Emily Grundy2 1
Department of Social Science, Health and Medicine, School of Social Science and Public Policy, King’s College London, London, UK 2 Department of Social Policy, London School of Economics and Political Science, London, UK Correspondence to Dr Giorgio Di Gessa, Department of Social Science, Health and Medicine, School of Social Science and Public Policy, King’s College London, The Strand, London WC2R 2LS, UK; [email protected] Received 3 May 2013 Revised 4 September 2013 Accepted 2 November 2013 Published Online First 23 November 2013
ABSTRACT Background ‘Active ageing’ has been promoted as a strategy for extending quality of life and healthy life expectancy. However, there is limited evidence from nationally representative longitudinal studies on whether engagement among older adults is associated with better outcomes and few studies have considered possible bias arising from differential study attrition. Methods We investigate associations between the engagement of people aged 50–69 in three types of activity with self-rated health and depression 2 years later using nationally representative longitudinal data from four European countries (Denmark, France, Italy and England). Data were drawn from waves 1 and 2 of the Survey of Health, Ageing and Retirement in Europe and the English Longitudinal Study of Ageing. Multivariable analysis was used to analyse associations between baseline activity and outcomes at follow-up controlling for socioeconomic, demographic and healthrelated variables at baseline. Multiple imputation techniques and sensitivity analyses were undertaken to investigate possible bias arising from sample attrition. Results Respondents in paid work at baseline were less likely to be depressed or to report poor or fair health at follow-up than those who were ‘inactive’, although not in Italy. Engagement in formal and informal activities was not significantly associated with health at follow-up. Sensitivity analyses showed that assuming that those in bad health were over-represented among study attritors weakened the association between work at baseline and health at follow-up. Conclusions Engagement in paid work may help maintain health in later life, although mechanisms and contextual influences need further investigation.
INTRODUCTION
To cite: Di Gessa G, Grundy E. J Epidemiol Community Health 2014;68:261–267.
The concept of ‘active ageing’ has been adopted as a policy priority in the European Union, which designated 2012 as the European Year for Active Ageing (http://europa.eu/ey2012/). The WHO defines ‘active ageing’ as continuing participation in any of a number of domains that extend beyond participation in economically productive activities to also include participation in cultural, spiritual and civic affairs, that contribute to personal wellbeing and quality of life, as well as to the well-being of other individuals and society at large.1 Engagement in social and productive activities has long been considered important for older people’s health and well-being.2–4 Previous studies generally support the idea that health is better among older people who are more active. For example, using longitudinal data drawn from the US Health and
Di Gessa G, et al. J Epidemiol Community Health 2014;68:261–267. doi:10.1136/jech-2013-202820
Retirement Study, Calvo5 found that older people in paid work had reduced morbidity and better perceived well-being when compared with those out of the labour market, even after controlling for baseline health. Similarly, in their systematic review of 16 US longitudinal studies, von Bonsdorff and Rantanen6 found that volunteering in old age was associated with increased life satisfaction, slower declines in self-reported health and better physical functioning, well-being and longevity. Bath and Deeg7 reviewed US and Scandinavian cross-sectional and longitudinal research that examined the impact of social participation on physical health outcomes and reported that educational, political and community-oriented activities were positively associated with better perceived health and reduced mortality. Comparative European work on engagement and health is still relatively sparse and generally considers few activities at a time, focusing mostly on paid work, caring for a sick person or volunteering.8–11 Moreover, much previous longitudinal research reports analyses of complete record datasets, without considering whether sample attrition might bias associations.12 13 In keeping with the WHO’s holistic conceptualisation of active ageing, in this paper, we investigate longitudinal associations between three types of engagement ( paid work, formal and informal activities) and two important indicators of health 2 years later in four European countries. Engagement of older people occurs within a socio-political context and may be influenced by the provision of services and generosity of benefits, pension schemes and labour, retirement and early retirement policies, as well as norms and values.14 15 It is well recognised that these contextual factors vary within Europe, and typologies have been developed that group together countries with similar policies.16 Although the fine details of these vary, most distinguish Nordic countries with generous welfare provisions (social democratic); countries that follow broadly Bismarckian welfare structure (such as France, Germany and Austria); those with liberal and mixed structures (such as the UK and Ireland, and the USA); and in more recent typologies, a Mediterranean group in which welfare policies are premised on greater familial exchange.16 The countries we selected for inclusion in the study were chosen to represent these four types of regime. The outcomes investigated were depression, which is associated with increased risk of coronary artery disease, cardiovascular death and worsened quality of life,17 18 and self-rated health (SRH), 261
Research report which is predictive of quality of life and mortality, even when physical health conditions are controlled for, and has the potential to capture positive dimensions of health rather than just presence of disease or disability.19 All models included control for baseline health, and multiple imputation (MI) techniques are used to examine the effects of attrition in the samples used.
in France and Italy where 34.9% and 28.9%, respectively, of respondents had dropped out of the survey by the second wave compared with 23.3% of Danish and 19.2% of English participants. As previous studies have shown,12 22 attrition is not random and tends to be higher among respondents of lower socioeconomic status and those in poorer health. Possible bias arising from attrition is considered in the final part of the paper.
METHODS Study population Data were drawn from two surveys, the English Longitudinal Study of Ageing (ELSA) and the multicountry Survey of Health, Ageing and Retirement in Europe (SHARE). Both are multidisciplinary longitudinal surveys of individuals aged 50 and over representative of the relevant national populations that were designed to enable comparative analyses. Specific details of sampling frames and methodology, weighting strategies and questionnaires have been reported elsewhere.20 21 We used data from Denmark, France, Italy and England, countries selected to represent different welfare regimes16 with good response and retention rates compared with the other available countries from the same regime type. Data were drawn from the first two waves of the surveys: the first wave of ELSA took place in 2002/2003, SHARE collected wave 1 data in 2004/2005; later waves were conducted biennially. We consider cohort respondents aged 50–69 at baseline, as only very small proportions of those in age groups older than this were in paid employment, and the analysis presented is necessarily restricted to those present in waves 1 and 2. Respondents who were missing baseline information for one or more variables in the analysis (n=402, 3.3%) were excluded. The proportion excluded for this reason ranged from 2.3% in Denmark to 4% in France. Respondents who had died by wave 2 (n=135, 1.1%) were also excluded. Initial sample sizes for the age groups of interest and attrition by wave 2 are shown in table 1. Attrition is defined as loss to follow-up for reasons other than death. Study drop-out was high, especially
Measures Outcomes Depression and SRH were measured using validated scales.23 24 ELSA included an abbreviated 8-item version of the Center for Epidemiologic Studies Depression Scale (CES-D),25 whereas SHARE used the EURO-D 12-item scale.24 Both asked respondents whether they had experienced any depressive symptoms, such as restless sleep or being unhappy in the week (ELSA) or month (SHARE) prior to interview. Those who reported three or more symptoms on the CES-D or four or more on the EURO-D scales were classified as being ‘depressed’.26 27 SRH in both SHARE and ELSA was measured in both waves using responses to a generic question (“Would you say your health is …”) on a five-point ordinal scale (excellent, very good, good, fair or poor). SRH responses may be sensitive to positioning of the question.28 In both studies, respondents were randomly allocated to rate their health status before or after a module of health questions. Since no systematic differences by socioeconomic or demographic characteristics were found, answers were combined. The five SRH items were dichotomised into ‘fair or poor’ versus better health in order to simplify the analysis. Although by dichotomising SRH we may lose some information, previous studies have shown that measures of morbidity and mortality are more strongly associated with adverse than with good SRH.23 Manderbacka et al29 have also shown that results using the dichotomised measure agree well with those based on treating the variable as continuous.
Table 1 Per cent distribution of sample members by activity, self-rated health and depression (people aged 50–69) in Denmark, France, Italy and England Denmark M No activities Paid work Formal engagement Informal engagement Health outcomes at baseline With depressive symptoms Self-rated health as poor or fair Baseline respondents (N) Health outcomes at follow-up With depressive symptoms Self-rated health as poor or fair Follow-up respondents (N) Initial response rate Drop out (excluding deaths) Dead between waves
France W
M
Italy W
M
England W
M
W
16.9 59.5 37.4 36.7
17.7 50.0 37.7 43.6
24.7 47.5 29.5 33.9
25.7 40.1 24.1 45.4
37.2 39.8 12.8 28.8
35.1 20.5 13.0 47.7
32.4 55.4 14.2 17.4
32.9 42.9 19.2 29.1
13.9 21.1 545
20.5 21.2 557
21.7 24.3 949
40.8 26.2 1049
21.7 27.2 777
38.0 36.2 999
19.0 24.2 3527
24.5 21.8 4037
13.4 22.1 413 63.2% 23.3% 1.5%
19.0 25.5 416
14.7 25.9 581 81.0% 34.9% 1.1%
20.6 33.2 535 54.5% 28.9% 1.0%
40.3 48.0 714
18.1 26.7 2792 67.0% 19.2% 1.0%
24.1 24.9 3257
40.4 31.0 692
Source: Denmark, France and Italy data obtained from Survey of Health, Ageing and Retirement in Europe (SHARE), 2004 and 2006; English data obtained from English Longitudinal Study of Ageing, 2002 and 2004. Data for England are based on the 8-point depression scale CES-D, and data for SHARE countries is based on the 12-item EURO-D measurement. Measures of engagement relate to past week in Denmark, France and Italy and past month in England. Own calculations; weighted data.
262
Di Gessa G, et al. J Epidemiol Community Health 2014;68:261–267. doi:10.1136/jech-2013-202820
Research report Variables on activity Respondents were classified as in paid work if they described their current situation as ‘employed or self-employed’ and if they were not ‘temporarily away from any work, including seasonal work’, to include only those with a current commitment to labour market activity. Participation in formal activities was defined as non-kin social activities linked to formalised associations, performed within an established structure with a regular schedule. Formal social involvement thus included organised voluntary work; attendance at training courses, and participation in political organisations, religious organisations or sport, social or other kinds of clubs. Informal and family-related engagement included activities with family members and/or friends, such as care provision for sick or disabled adults, provision of help to family, friends or neighbours and others from inside and outside the household; and looking after grandchildren without the presence of their parents. SHARE asked about participation in these activities almost every week or more often, as well as less frequent engagement. We chose the indicator of weekly activities as this was most suitable to the research hypothesis. However, this was at some cost in terms of comparability as ELSA only collected data on engagement in the month prior to interview.
Other covariates On the basis of the existing literature we identified characteristics known to be associated with either or both health and activity and accordingly controlled for the following characteristics at baseline: age, gender, education, wealth, living alone and functional limitation. Educational qualifications were re-coded to three categories representing low, mid and high education using the International Standard Classification of Education classification.30 Wealth was measured using quintile of the harmonised sum of the net value of properties, non-housing financial wealth and business assets created by the RAND Corporation (http:// www.mmicdata.rand.org/meta/). Living arrangements were measured using a dichotomised indicator of whether the respondent lived alone or with others. Functional health was measured using a dichotomised variable indicating whether or not respondents had any limitations activities of daily living (ADL); Instrumental Activities of Daily Living (IADL) or physical performance activities using the Nagi item battery (NAGI).31
Statistical analyses Preliminary analyses were carried out separately for men and women in each country but given the similar patterns observed for men and women; results for both genders combined are presented here. Analyses were first undertaken for participants with complete data on all variables examined. In a second stage, MI under the Missing At Random (MAR) assumption (ie, missingness in a variable is independent of the missing values themselves after conditioning on the observed data)32 was used to explore the effects of missing data. In this analysis, SRH and depression at follow-up were imputed separately by country and gender using chained equations. All baseline covariates used in the analysis models were included in the imputation of both outcomes (ie, age, gender, whether the respondent lived alone or not, ADL, IADL, NAGI, SRH, depression, education, wealth, engagement in the various activities). Additionally, number of chronic diseases and marital status were included as auxiliary variables in the MI model as both are predictors of health outcomes in Western populations. As previously discussed, the very small proportions of respondents with missing information on
baseline covariates were excluded. The chained equation process was continued for 20 cycles, and 200 imputed datasets were created. The results of analyses on each individual dataset were combined using Rubin’s rules.32 Pattern mixture models33 were used to assess whether and how much various plausible ‘arbitrary’ assumptions about the missing data mechanism affected the results. The robustness of the results was tested by doing successive analyses assuming that among drop-outs the proportion of people who had depressive symptoms and who rated their health as poor or fair was increased by 20% and 33%. All analyses were performed using Stata, V.12.34
RESULTS Descriptive statistics Table 1 shows the level of engagement and health of older respondents by gender and country. There were large variations by country, with the highest levels of activity in Denmark and the lowest in Italy. There were also differences by country and gender in the proportions with depressive symptoms or poor or fair SRH. Danish and English respondents had better health than Italian and French respondents both at baseline and at follow-up.
Associations between engagement at baseline and health indicators at follow-up Depression Table 2 shows results from logistic regression models that investigated associations between baseline engagement and depression at follow-up in each country considered, controlling for baseline socioeconomic and demographic characteristics and health. In England, engagement in paid work at baseline was significantly associated with lower odds of depression at follow-up; in Italy and Denmark, those in paid work also had lower odds of depression, but the results were of marginal statistical significance ( p
The relationship between active ageing and health using longitudinal data from Denmark, France, Italy and England Giorgio Di Gessa,1 Emily Grundy2 1
Department of Social Science, Health and Medicine, School of Social Science and Public Policy, King’s College London, London, UK 2 Department of Social Policy, London School of Economics and Political Science, London, UK Correspondence to Dr Giorgio Di Gessa, Department of Social Science, Health and Medicine, School of Social Science and Public Policy, King’s College London, The Strand, London WC2R 2LS, UK; [email protected] Received 3 May 2013 Revised 4 September 2013 Accepted 2 November 2013 Published Online First 23 November 2013
ABSTRACT Background ‘Active ageing’ has been promoted as a strategy for extending quality of life and healthy life expectancy. However, there is limited evidence from nationally representative longitudinal studies on whether engagement among older adults is associated with better outcomes and few studies have considered possible bias arising from differential study attrition. Methods We investigate associations between the engagement of people aged 50–69 in three types of activity with self-rated health and depression 2 years later using nationally representative longitudinal data from four European countries (Denmark, France, Italy and England). Data were drawn from waves 1 and 2 of the Survey of Health, Ageing and Retirement in Europe and the English Longitudinal Study of Ageing. Multivariable analysis was used to analyse associations between baseline activity and outcomes at follow-up controlling for socioeconomic, demographic and healthrelated variables at baseline. Multiple imputation techniques and sensitivity analyses were undertaken to investigate possible bias arising from sample attrition. Results Respondents in paid work at baseline were less likely to be depressed or to report poor or fair health at follow-up than those who were ‘inactive’, although not in Italy. Engagement in formal and informal activities was not significantly associated with health at follow-up. Sensitivity analyses showed that assuming that those in bad health were over-represented among study attritors weakened the association between work at baseline and health at follow-up. Conclusions Engagement in paid work may help maintain health in later life, although mechanisms and contextual influences need further investigation.
INTRODUCTION
To cite: Di Gessa G, Grundy E. J Epidemiol Community Health 2014;68:261–267.
The concept of ‘active ageing’ has been adopted as a policy priority in the European Union, which designated 2012 as the European Year for Active Ageing (http://europa.eu/ey2012/). The WHO defines ‘active ageing’ as continuing participation in any of a number of domains that extend beyond participation in economically productive activities to also include participation in cultural, spiritual and civic affairs, that contribute to personal wellbeing and quality of life, as well as to the well-being of other individuals and society at large.1 Engagement in social and productive activities has long been considered important for older people’s health and well-being.2–4 Previous studies generally support the idea that health is better among older people who are more active. For example, using longitudinal data drawn from the US Health and
Di Gessa G, et al. J Epidemiol Community Health 2014;68:261–267. doi:10.1136/jech-2013-202820
Retirement Study, Calvo5 found that older people in paid work had reduced morbidity and better perceived well-being when compared with those out of the labour market, even after controlling for baseline health. Similarly, in their systematic review of 16 US longitudinal studies, von Bonsdorff and Rantanen6 found that volunteering in old age was associated with increased life satisfaction, slower declines in self-reported health and better physical functioning, well-being and longevity. Bath and Deeg7 reviewed US and Scandinavian cross-sectional and longitudinal research that examined the impact of social participation on physical health outcomes and reported that educational, political and community-oriented activities were positively associated with better perceived health and reduced mortality. Comparative European work on engagement and health is still relatively sparse and generally considers few activities at a time, focusing mostly on paid work, caring for a sick person or volunteering.8–11 Moreover, much previous longitudinal research reports analyses of complete record datasets, without considering whether sample attrition might bias associations.12 13 In keeping with the WHO’s holistic conceptualisation of active ageing, in this paper, we investigate longitudinal associations between three types of engagement ( paid work, formal and informal activities) and two important indicators of health 2 years later in four European countries. Engagement of older people occurs within a socio-political context and may be influenced by the provision of services and generosity of benefits, pension schemes and labour, retirement and early retirement policies, as well as norms and values.14 15 It is well recognised that these contextual factors vary within Europe, and typologies have been developed that group together countries with similar policies.16 Although the fine details of these vary, most distinguish Nordic countries with generous welfare provisions (social democratic); countries that follow broadly Bismarckian welfare structure (such as France, Germany and Austria); those with liberal and mixed structures (such as the UK and Ireland, and the USA); and in more recent typologies, a Mediterranean group in which welfare policies are premised on greater familial exchange.16 The countries we selected for inclusion in the study were chosen to represent these four types of regime. The outcomes investigated were depression, which is associated with increased risk of coronary artery disease, cardiovascular death and worsened quality of life,17 18 and self-rated health (SRH), 261
Research report which is predictive of quality of life and mortality, even when physical health conditions are controlled for, and has the potential to capture positive dimensions of health rather than just presence of disease or disability.19 All models included control for baseline health, and multiple imputation (MI) techniques are used to examine the effects of attrition in the samples used.
in France and Italy where 34.9% and 28.9%, respectively, of respondents had dropped out of the survey by the second wave compared with 23.3% of Danish and 19.2% of English participants. As previous studies have shown,12 22 attrition is not random and tends to be higher among respondents of lower socioeconomic status and those in poorer health. Possible bias arising from attrition is considered in the final part of the paper.
METHODS Study population Data were drawn from two surveys, the English Longitudinal Study of Ageing (ELSA) and the multicountry Survey of Health, Ageing and Retirement in Europe (SHARE). Both are multidisciplinary longitudinal surveys of individuals aged 50 and over representative of the relevant national populations that were designed to enable comparative analyses. Specific details of sampling frames and methodology, weighting strategies and questionnaires have been reported elsewhere.20 21 We used data from Denmark, France, Italy and England, countries selected to represent different welfare regimes16 with good response and retention rates compared with the other available countries from the same regime type. Data were drawn from the first two waves of the surveys: the first wave of ELSA took place in 2002/2003, SHARE collected wave 1 data in 2004/2005; later waves were conducted biennially. We consider cohort respondents aged 50–69 at baseline, as only very small proportions of those in age groups older than this were in paid employment, and the analysis presented is necessarily restricted to those present in waves 1 and 2. Respondents who were missing baseline information for one or more variables in the analysis (n=402, 3.3%) were excluded. The proportion excluded for this reason ranged from 2.3% in Denmark to 4% in France. Respondents who had died by wave 2 (n=135, 1.1%) were also excluded. Initial sample sizes for the age groups of interest and attrition by wave 2 are shown in table 1. Attrition is defined as loss to follow-up for reasons other than death. Study drop-out was high, especially
Measures Outcomes Depression and SRH were measured using validated scales.23 24 ELSA included an abbreviated 8-item version of the Center for Epidemiologic Studies Depression Scale (CES-D),25 whereas SHARE used the EURO-D 12-item scale.24 Both asked respondents whether they had experienced any depressive symptoms, such as restless sleep or being unhappy in the week (ELSA) or month (SHARE) prior to interview. Those who reported three or more symptoms on the CES-D or four or more on the EURO-D scales were classified as being ‘depressed’.26 27 SRH in both SHARE and ELSA was measured in both waves using responses to a generic question (“Would you say your health is …”) on a five-point ordinal scale (excellent, very good, good, fair or poor). SRH responses may be sensitive to positioning of the question.28 In both studies, respondents were randomly allocated to rate their health status before or after a module of health questions. Since no systematic differences by socioeconomic or demographic characteristics were found, answers were combined. The five SRH items were dichotomised into ‘fair or poor’ versus better health in order to simplify the analysis. Although by dichotomising SRH we may lose some information, previous studies have shown that measures of morbidity and mortality are more strongly associated with adverse than with good SRH.23 Manderbacka et al29 have also shown that results using the dichotomised measure agree well with those based on treating the variable as continuous.
Table 1 Per cent distribution of sample members by activity, self-rated health and depression (people aged 50–69) in Denmark, France, Italy and England Denmark M No activities Paid work Formal engagement Informal engagement Health outcomes at baseline With depressive symptoms Self-rated health as poor or fair Baseline respondents (N) Health outcomes at follow-up With depressive symptoms Self-rated health as poor or fair Follow-up respondents (N) Initial response rate Drop out (excluding deaths) Dead between waves
France W
M
Italy W
M
England W
M
W
16.9 59.5 37.4 36.7
17.7 50.0 37.7 43.6
24.7 47.5 29.5 33.9
25.7 40.1 24.1 45.4
37.2 39.8 12.8 28.8
35.1 20.5 13.0 47.7
32.4 55.4 14.2 17.4
32.9 42.9 19.2 29.1
13.9 21.1 545
20.5 21.2 557
21.7 24.3 949
40.8 26.2 1049
21.7 27.2 777
38.0 36.2 999
19.0 24.2 3527
24.5 21.8 4037
13.4 22.1 413 63.2% 23.3% 1.5%
19.0 25.5 416
14.7 25.9 581 81.0% 34.9% 1.1%
20.6 33.2 535 54.5% 28.9% 1.0%
40.3 48.0 714
18.1 26.7 2792 67.0% 19.2% 1.0%
24.1 24.9 3257
40.4 31.0 692
Source: Denmark, France and Italy data obtained from Survey of Health, Ageing and Retirement in Europe (SHARE), 2004 and 2006; English data obtained from English Longitudinal Study of Ageing, 2002 and 2004. Data for England are based on the 8-point depression scale CES-D, and data for SHARE countries is based on the 12-item EURO-D measurement. Measures of engagement relate to past week in Denmark, France and Italy and past month in England. Own calculations; weighted data.
262
Di Gessa G, et al. J Epidemiol Community Health 2014;68:261–267. doi:10.1136/jech-2013-202820
Research report Variables on activity Respondents were classified as in paid work if they described their current situation as ‘employed or self-employed’ and if they were not ‘temporarily away from any work, including seasonal work’, to include only those with a current commitment to labour market activity. Participation in formal activities was defined as non-kin social activities linked to formalised associations, performed within an established structure with a regular schedule. Formal social involvement thus included organised voluntary work; attendance at training courses, and participation in political organisations, religious organisations or sport, social or other kinds of clubs. Informal and family-related engagement included activities with family members and/or friends, such as care provision for sick or disabled adults, provision of help to family, friends or neighbours and others from inside and outside the household; and looking after grandchildren without the presence of their parents. SHARE asked about participation in these activities almost every week or more often, as well as less frequent engagement. We chose the indicator of weekly activities as this was most suitable to the research hypothesis. However, this was at some cost in terms of comparability as ELSA only collected data on engagement in the month prior to interview.
Other covariates On the basis of the existing literature we identified characteristics known to be associated with either or both health and activity and accordingly controlled for the following characteristics at baseline: age, gender, education, wealth, living alone and functional limitation. Educational qualifications were re-coded to three categories representing low, mid and high education using the International Standard Classification of Education classification.30 Wealth was measured using quintile of the harmonised sum of the net value of properties, non-housing financial wealth and business assets created by the RAND Corporation (http:// www.mmicdata.rand.org/meta/). Living arrangements were measured using a dichotomised indicator of whether the respondent lived alone or with others. Functional health was measured using a dichotomised variable indicating whether or not respondents had any limitations activities of daily living (ADL); Instrumental Activities of Daily Living (IADL) or physical performance activities using the Nagi item battery (NAGI).31
Statistical analyses Preliminary analyses were carried out separately for men and women in each country but given the similar patterns observed for men and women; results for both genders combined are presented here. Analyses were first undertaken for participants with complete data on all variables examined. In a second stage, MI under the Missing At Random (MAR) assumption (ie, missingness in a variable is independent of the missing values themselves after conditioning on the observed data)32 was used to explore the effects of missing data. In this analysis, SRH and depression at follow-up were imputed separately by country and gender using chained equations. All baseline covariates used in the analysis models were included in the imputation of both outcomes (ie, age, gender, whether the respondent lived alone or not, ADL, IADL, NAGI, SRH, depression, education, wealth, engagement in the various activities). Additionally, number of chronic diseases and marital status were included as auxiliary variables in the MI model as both are predictors of health outcomes in Western populations. As previously discussed, the very small proportions of respondents with missing information on
baseline covariates were excluded. The chained equation process was continued for 20 cycles, and 200 imputed datasets were created. The results of analyses on each individual dataset were combined using Rubin’s rules.32 Pattern mixture models33 were used to assess whether and how much various plausible ‘arbitrary’ assumptions about the missing data mechanism affected the results. The robustness of the results was tested by doing successive analyses assuming that among drop-outs the proportion of people who had depressive symptoms and who rated their health as poor or fair was increased by 20% and 33%. All analyses were performed using Stata, V.12.34
RESULTS Descriptive statistics Table 1 shows the level of engagement and health of older respondents by gender and country. There were large variations by country, with the highest levels of activity in Denmark and the lowest in Italy. There were also differences by country and gender in the proportions with depressive symptoms or poor or fair SRH. Danish and English respondents had better health than Italian and French respondents both at baseline and at follow-up.
Associations between engagement at baseline and health indicators at follow-up Depression Table 2 shows results from logistic regression models that investigated associations between baseline engagement and depression at follow-up in each country considered, controlling for baseline socioeconomic and demographic characteristics and health. In England, engagement in paid work at baseline was significantly associated with lower odds of depression at follow-up; in Italy and Denmark, those in paid work also had lower odds of depression, but the results were of marginal statistical significance ( p
